To evaluate whether the genetic signal of depression was independent of signals from comorbid conditions, we employed multi-trait-based conditional and joint analysis (mtCOJO) in GCTA.48 With mtCOJO, per-SNP effect estimates and association statistics of MDD-META were adjusted for the causal effects between MDD and seven comorbid conditions estimated by Mendelian randomization. We required at least 2 genome-wide significant SNPs after Heidi outlier testing with which to estimate causality between phenotypes. MDD was conditioned eight times: once each for alcohol use disorder, digestive disorders, educational attainment, fibromyalgia, neuroticism (SSGAC), schizophrenia, and subjective well-being; and once using all seven correlates simultaneously. In this experimental design, we generated 8 new versions of depression GWAS summary statistics termed “conditioned” GWAS to analyze for heritability, genetic correlation versus the original unconditioned depression GWAS, SNP effects, and p-value survival. These analyses are described in Methods under Post-GWAS analysis: Linkage Disequilibrium Score Regression. Conditioned GWAS generated from mtCOJO are free of collider biases when estimate causal relationship between depression and each comorbid condition.49
